nmtoan91 / lshkrepresentatives Goto Github PK

Hi,

Again, thanks for the great algorithm!

I just have a question about instances where there's a mix of ordinal and binary data. Does the algorithm handle all data as nominal, or is it possible for it to handle ordinal and binary/nominal data differently? The thing is that it appears to me, when running with mixed ordinal and binary data, that the algorithm weighs binary variables more (or at least distinguishes them more clearly between clusters), whereas ordinal variables (even though knowing the particular data I've been running the algorithm on there should naturally be groups which are strongly towards either end of the "spectrum") have relatively similar value proportions.

Best regards,
Daniil

Hi, first thank you for this work. I would like use your code in my experiments. However, it seems that the DILCA dissimilarity is not used in distance computations in your current implementation of LSHkRepresentatives. Am I wrong?

Hi,

Thank you for this clustering method, it is very useful. However, one thing that I think would provide a lot of benefit, is if there was a function, similar to the predict() function in the scikit-learn library, with which one may assign cluster labels to new/old data on a previously trained LSH-k-representatives instance.

Best regards,
Daniil